1. Field of the Invention
The present invention relates to a method and device for evaluating the noise associated with turbocodes, and to systems using them.
2. Related Art
Turbocodes are very advantageous in conditions of low signal to noise ratios (SNRs). A conventional turbo-encoder consists of two recursive systematic convolutional (RSC) encoders and an interleaver, disposed as shown in FIG. 1. The turbo-encoder supplies as an output three series of binary elements (x, y1, y2), where x is the so-called systematic output of the turbo-encoder, that is to say one which has not undergone any processing with respect to the input signal x, y1 is the output encoded by the first RSC encoder, and y2 is the output encoded by the second RSC encoder after passing through the interleaver.
For more details on turbocodes, reference could usefully be made to the article by C. Berrou, A. Glavieux and P. Thitimajshima entitled “Near Shannon limit error-correcting coding and decoding: turbo-codes”, ICC '93, Geneva.
FIG. 2 depicts an example of a conventional turbo-decoder able to decode data supplied by a turbo-encoder like the one in FIG. 1. The inputs x′, y1′, y2′ of the turbodecoder are the outputs of the turbo-encoder altered by the transmission channel and the transmission and reception processes. The structure of such a turbodecoder is well known to persons skilled in the art and will therefore not be described in detail here.
It requires in particular two decoders, referred to as “Decoder 1” and “Decoder 2” in FIG. 2, for example of the BCJR type, that is to say using the Bahl, Cocke, Jelinek and Raviv algorithm, or of the SOVA (“Soft Output Viterbi Algorithm”) type. The data supplied as an input to the decoders 1 and 2 take into account the signal to noise ratio.
A conventional turbodecoder also requires a looping back of the output of the deinterleaver π2 onto the input of the first decoder, in order to transmit the so-called “extrinsic” information from the second decoder to the first decoder.
It can be shown that the result of the decoding depends on the noise impairing the transmission channel as well as the transmission and reception processes.
In a theoretical study or during a simulation, the “noise” parameter is generally a given in the problem.
On the other hand, in an application to the real world, the noise is a characteristic related to the channel and can vary from one data transmission to another, or even during the same data transmission. In fact, it is possible to know only approximate statistics of the noise.
In the case of turbocodes, the decoding system can work with a highly incorrect noise estimation; nevertheless, its ability to correct the errors which have been introduced by the channel noise will be decreased. In this case, the encoding cost, which is the redundancy of the data, is high compared with the gain in performance and makes the system inadequate.
Different techniques are known for effecting a statistical evaluation of the noise.
For example, since the appearance of data transmission modems, the possibility has been known of using the constellation of the signal for establishing statistics of the noise. Thus, when the points transmitted are coded in accordance with an NRZ (Non-Return to Zero) modulation, the original symbols belong to a set {−1 ; +1}. It is assumed that the symbols received have undergone a white Gaussian noise addition; they are therefore distributed in accordance with a distribution illustrated in FIG. 3.
It is then possible to extract an estimation of the noise which, in the case illustrated in FIG. 3, corresponds to the standard deviation of the symbols received with respect to a mean centered on the position of the symbols sent. It should be noted that the centering on the theoretical symbols is a consequence of using a white Gaussian noise.
During an actual transmission, the assumption according to which the noise on the channel is Gaussian is an approximation. It is however possible to obtain an estimation of the noise by accumulating the measurements of the deviations of each symbol received with respect to a theoretical symbol situated at the shortest distance from this received symbol, and then dividing this accumulation by the total number of symbols received. Here a maximum likelihood criterion is applied, which assimilates a received symbol to its closest neighbour. Thus the evaluation of the noise B on a modulation, whether it is plotted on the Fresnel plane with one or two dimensions, is effected by means of the following operation:   B  =                    ∑                  i          =          1                N            ⁢                        min          i                ⁢                  [                                    (                              ·                                  -                                      S                    i                                                              )                        2                    ]                      N  where i is an integer, N designates the number of symbols per frame or block, “min” designates the smallest Euclidian distance between a received symbol and the theoretical points of the constellation, the sign • designates the position of the received symbols and Si designates the positions of the theoretical symbols.
This technique is known notably in the field of modems, where it is used for obtaining a negotiation of the bit rate per symbol as a function of the state of the channel.
However, this solution has the drawback of introducing an inaccuracy, since the noise suffered by the original symbol can be such that the received symbol is situated at a smaller distance from a theoretical symbol different from the original symbol than the original symbol itself.